Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75565.71748585007,
-      "required_pages_per_second_upper_estimate": 0.0065498448721134726
+      "required_token_positions_per_second": 74662.79907872532,
+      "required_pages_per_second_upper_estimate": 0.006471582193009681
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9445.714685731258,
-      "required_pages_per_second_upper_estimate": 0.0037680368141579934
+      "required_token_positions_per_second": 9332.849884840665,
+      "required_pages_per_second_upper_estimate": 0.003723013357603584
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18891.429371462516,
-      "required_pages_per_second_upper_estimate": 0.004370199707286442
+      "required_token_positions_per_second": 18665.69976968133,
+      "required_pages_per_second_upper_estimate": 0.0043179811366196716
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28337.144057193775,
-      "required_pages_per_second_upper_estimate": 0.0017255833995783267
+      "required_token_positions_per_second": 27998.549654521994,
+      "required_pages_per_second_upper_estimate": 0.0017049647769231534
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28337.144057193775,
-      "required_pages_per_second_upper_estimate": 0.0021814717917580536
+      "required_token_positions_per_second": 27998.549654521994,
+      "required_pages_per_second_upper_estimate": 0.0021554058573510842
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28337.144057193775,
-      "required_pages_per_second_upper_estimate": 0.35185249086995757
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27998.549654521994,
+      "required_pages_per_second_upper_estimate": 0.34764828159134303
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.0962717990702231,
-    "producer_p95_seconds": 0.38405086399870925,
-    "gpu_update_seconds": 0.9756805394431121,
-    "producer_batches_per_second": 10.387257843499679,
-    "gpu_batches_per_second": 1.0249256386427146,
-    "producer_headroom": 10.134645336080464,
+    "producer_batch_seconds": 0.1735889495543006,
+    "producer_p95_seconds": 0.4106779829962761,
+    "gpu_update_seconds": 0.9874797209552825,
+    "producer_batches_per_second": 5.760735361136502,
+    "gpu_batches_per_second": 1.0126790239627457,
+    "producer_headroom": 5.688609346912302,
     "sustainable": true,
-    "buffer_seconds": 60.49219344547295,
+    "buffer_seconds": 60.23626297827223,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 242955.97611479394,
-    "upload_bytes_per_second": 5735929.90473248,
+    "generation_bytes_per_second": 263058.4646103507,
+    "upload_bytes_per_second": 6032254.231295674,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.1326967485701,
+    "estimated_full_shards_per_hour": 56.446222817734636,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 4.655135680000967,
-      "bytes": 7374926,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.311550709000585,
       "bytes": 11722471,
@@ -562,6 +556,12 @@
       "bytes": 6815769,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1106542109992006,
+      "bytes": 7819068,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -366,3 +366,4 @@
 {"step": 151600, "loss": 0.5758757181465626, "lr": 0.0003, "grad_norm": 0.16094720975034807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22186405283, "tokens_per_second": 151403.1334321257, "data_wait_s": 9.796000085771084e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 151700, "loss": 0.6355745941400528, "lr": 0.0003, "grad_norm": 0.1451971084639051, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22201131978, "tokens_per_second": 149029.01259873327, "data_wait_s": 9.254996257368475e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 151800, "loss": 0.6344895455986261, "lr": 0.0003, "grad_norm": 0.16112494940064362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22215780946, "tokens_per_second": 149172.94995248725, "data_wait_s": 1.5428995538968593e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 151900, "loss": 0.6639365497976542, "lr": 0.0003, "grad_norm": 0.14157858284681954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22230487947, "tokens_per_second": 128559.01434671022, "data_wait_s": 1.8451006326358765e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbd17596ecec793ccf730dcb4a4ec01c1dcd7a9fc2740a48620064f356f4c0af
-size 785070
+oid sha256:b6f24753252e58d7a95a5b84edf6c7ac3d25cdc549551a992744710feba38566
+size 787188